Method and device for optical biopsy of biological tissue

Abstract

The invention relates to the technical field of biological tissue biopsy, in particular to a biological tissue optical biopsy method and device.The method comprises the steps that a two-dimensional cross section image or a three-dimensional image of a target lumen channel is obtained; and converting the two-dimensional cross section image or the three-dimensional image through a pre-constructed cyclic generative adversarial network model based on contrast learning assistance to generate a histopathological image of the target lumen. Therefore, the problems that existing endoscopic examination has certain limitation, whether a lesion area exists or not is difficult to accurately judge through tissue images, pathological examination needs to be matched, body injuries are easily brought to patients and the like are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of biological tissue biopsy, and particularly relates to a method and device for optical biopsy of biological tissues. Background Art

[0002] In the field of early lung cancer screening, the existing technologies mainly include low-dose computed tomography (LDCT), bronchoscopy, and tissue biopsy, and each of these technologies has certain limitations.

[0003] Low-dose computed tomography (LDCT), as the most studied and widely recognized preferred method for early lung cancer screening at present, has a relatively low radiation dose and can obtain relatively clear lung scan images. Its principle is to scan a certain thickness of lung tissue with an X-ray beam, and the detector receives the X-rays passing through the tissue, and images are formed based on the absorption differences of the tissue for X-rays. The imaging accuracy of LDCT only stays at the tissue level, and it can only present the general anatomical structure of the lungs. For early lung nodules with small diameters and atypical shapes, it is difficult to accurately judge their nature, which easily leads to missed diagnosis or misdiagnosis, causing patients to miss the best treatment opportunity.

[0004] Bronchoscopy includes various methods such as fiberoptic bronchoscopy (BFS) examination, autofluorescence bronchoscopy (AFB) examination, endobronchial ultrasound (EBUS) examination, etc. Fiberoptic bronchoscopy is a flexible and bendable endoscope that is inserted into the trachea and bronchi through the nasal cavity or oral cavity. The glass fiber bundle inside the endoscope body is used to introduce an external cold light source to illuminate the inside of the airway, and doctors can directly observe the mucosal morphology of the trachea and bronchi through this. However, since the lesion is only a few layers of cells (0.2 - 1 mm), it is difficult to judge abnormalities with the naked eye. Autofluorescence bronchoscopy is based on the spontaneous fluorescence characteristics of human tissues. By using the differences in fluorescence intensity and spectrum between normal tissues and tumor tissues under the excitation of a laser with a specific wavelength, lesions are identified by comparing the fluorescence regions. However, this method cannot be qualitatively determined alone and has limited accuracy. EBUS installs an ultrasonic probe at the front end of the bronchoscope and uses the echo reflection principle of ultrasound to clearly display the hierarchical structure of the tracheal and bronchial walls and the tissue conditions of the surrounding mediastinum, hilar lymph nodes, etc. However, these endoscopic examination technologies all have certain limitations, and ultimately often need to be combined with tissue biopsy for accurate diagnosis. For example, although EBUS combined with puncture biopsy is relatively accurate in judging the relationship between the lesion and the surrounding area and staging, it still cannot avoid the risks brought by puncture; BFS relies on biopsy and is greatly affected by the material collection. The quality of the sample obtained is directly related to the accuracy of the diagnosis result; when AFB is used alone, it is difficult to accurately judge the nature of the lesion and there is a possibility of missed diagnosis.

[0005] The "gold standard" for lung cancer diagnosis is tissue biopsy, which involves obtaining a sample of the diseased tissue for pathological examination to determine the nature of the lesion. In the early stage of lung cancer, when small nodules are detected in the lungs through imaging examinations, tissue biopsy is required to sample the suspicious nodules in order to detect early lung cancer in a timely manner. However, tissue biopsy is an invasive procedure with many risks. During percutaneous biopsy, the puncture needle entering the human tissue may damage blood vessels, resulting in symptoms such as hemoptysis; open biopsy surgery is prone to infection, and after lung tissue biopsy, it may also lead to lung infection, and in severe cases, even cause complications and death, bringing great trauma to the patient's body and affecting the patient's subsequent treatment and rehabilitation to a certain extent.

[0006] Therefore, there is an urgent need for a method that does not require puncturing the patient, which can avoid the physical trauma and potential risks caused by puncture and provide a safer screening option for patients. Summary of the Invention

[0007] The present invention provides a method and device for optical biopsy of biological tissues to solve the problems that existing bronchoscopic examinations have certain limitations, it is difficult to accurately determine whether there are diseased areas through tissue images, and it needs to be combined with pathological examinations, which is likely to cause physical trauma to patients.

[0008] In a first aspect embodiment of the present invention, a method for optical biopsy of biological tissues is provided, including the following steps: obtaining a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of a target lumen based on endoscopic optical coherence tomography technology; converting the two-dimensional cross-sectional image or the three-dimensional stereoscopic image through a pre-constructed cycle generative adversarial network model assisted by contrast learning to generate a tissue pathological image of the target lumen.

[0009] Optionally, the obtaining a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of a target lumen based on endoscopic optical coherence tomography technology includes:

[0010] Based on endoscopic optical coherence tomography technology, an endoscope is set on a preset frequency-domain optical coherence tomography device;

[0011] Using the preset frequency-domain optical coherence tomography device to emit visible light or near-infrared light, and scanning the lumen tissue of the target lumen through the endoscope to obtain the two-dimensional cross-sectional image or the three-dimensional stereoscopic image.

[0012] Optionally, the endoscope is provided with a graded-index lens or a metasurface lens composed of an integrated scatterer array to converge visible light or near-infrared light onto the lumen tissue of the target lumen.

[0013] Optionally, the two-dimensional cross-sectional image or the three-dimensional stereoscopic image is transformed by a pre-constructed cycle generative adversarial network model assisted by contrast learning to generate the histopathological image of the target lumen, including:

[0014] Training the cycle generative adversarial network model assisted by contrast learning with a preset unpaired dataset, and using the cycle consistency loss to relieve the dependence on paired training data to obtain an image local feature contrast learning enhanced cycle generative adversarial network model;

[0015] Using the image local feature contrast learning enhanced cycle generative adversarial network model to transform the two-dimensional cross-sectional image or the three-dimensional stereoscopic image to generate the histopathological image of the target lumen.

[0016] Optionally, the preset unpaired dataset includes two-dimensional cross-sectional images and three-dimensional stereoscopic images of the target lumens of different individuals, or three-dimensional stereoscopic images and histopathological images.

[0017] Optionally, the preset image segmentation model adopts at least one of a Transformer converter model, a U-Net deep learning model, and a convolutional neural network model.

[0018] Optionally, the expression of the overall loss function of the cycle generative adversarial network model assisted by contrast learning is:

[0019] L total (G H&E ,G OCT ,D H&E ,D OCT )=L GAN (G H&E ,D H&E )+L GAN (G OCT ,D OCT )

[0020] +λ·L cyc (G H&E ,G OCT )+α·L cl (G H&E ,G OCT )+β·L perceptual (G H&E ,G OCT )

[0021] Among them, L total is the overall loss function, G H&E is the histopathological image generator, G OCT is the optical coherence tomography generator, D H&Eis a histopathological image discriminator, D OCT is an optical coherence tomography discriminator, L GAN is an adversarial loss function, L cyc is a cycle consistency loss function, L cl is an image feature contrast loss function, L perceptual is a perceptual loss function, and λ, α, and β represent coefficients for controlling the correlation of different losses.

[0022] An embodiment of the second aspect of the present invention provides a device for optical biopsy of biological tissues, including: an image acquisition module for acquiring a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of a target lumen based on endoscopic optical coherence tomography technology; an image conversion module for converting the two-dimensional cross-sectional image or the three-dimensional stereoscopic image through a pre-constructed cycle generative adversarial network model assisted by contrast learning to generate a histopathological image of the target lumen.

[0023] An embodiment of the third aspect of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the program to implement the method for optical biopsy of biological tissues as described in the above embodiment.

[0024] An embodiment of the fourth aspect of the present invention provides a computer program product, and when the computer program / instructions are executed by a processor, the method for optical biopsy of biological tissues as above is implemented.

[0025] An embodiment of the fifth aspect of the present invention provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and when the program is executed by a processor, the method for optical biopsy of biological tissues as above is implemented.

[0026] The method and device for optical biopsy of biological tissues proposed in the embodiments of the present invention perform imaging analysis on luminal tissues through optical coherence tomography technology, and generate corresponding tissue biopsy images with the aid of a cycle generative adversarial network model assisted by contrastive learning. There is no need to puncture the patient, avoiding physical trauma and potential risks caused by puncture, and providing a safer screening option for patients; in terms of clinical application, once this technology is widely promoted, it will have an important impact on the early screening of lung cancer. Early detection of lung cancer is crucial for improving the survival rate and prognosis of patients; through a long-distance endoscopic optical coherence tomography system, early lung cancer can be detected more accurately, enabling patients to receive treatment at an early stage of the disease, greatly improving the treatment success rate and the quality of life of patients; from the perspective of the development of the medical industry, the application of this technology will promote the innovation of lung cancer screening technology and prompt the research and development of medical equipment to develop in a more accurate and non-invasive direction; from a social perspective, it can effectively reduce the heavy economic burden brought by lung cancer treatment to families and society, improve the overall health level and quality of life of society, and has significant social benefits.

[0027] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The above-mentioned and / or additional aspects and advantages of the present invention will become apparent and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0029] Figure 1 is a flowchart of a method for optical biopsy of biological tissues provided by an embodiment of the present invention;

[0030] Figure 2 is a schematic structural diagram of a swept-source optical coherence tomography device provided by an embodiment of the present invention;

[0031] Figure 3 is a schematic structural diagram of a spectral-domain optical coherence tomography device provided by an embodiment of the present invention;

[0032] Figure 4 is a schematic structural diagram of a lateral distal scanning probe based on an objective lens provided by an embodiment of the present invention;

[0033] Figure 5 is a schematic structural diagram of a lateral distal scanning probe composed of fused optical fibers provided by an embodiment of the present invention;

[0034] Figure 6 is a schematic structural diagram of a gradient-index lens provided by an embodiment of the present invention;

[0035] Figure 7Schematic diagram of the metasurface lens structure composed of sub-wavelength interval scatterer arrays provided by the embodiments of the present invention;

[0036] Figure 8 Schematic diagram of the paired and unpaired data sets provided by the embodiments of the present invention;

[0037] Figure 9 Schematic diagram of the overall framework of the looped structure of the contrastive learning-assisted cycle generative adversarial network model provided by the embodiments of the present invention;

[0038] Figure 10 Schematic diagram of the effect of the residual block acting on the generator network provided by the embodiments of the present invention;

[0039] Figure 11 Schematic diagram of the generative adversarial network and the cycle consistency loss provided by the embodiments of the present invention;

[0040] Figure 12 Structure diagram of the pre-trained VGG model provided by the embodiments of the present invention;

[0041] Figure 13 Structure diagram of the contrastive learning model provided by the embodiments of the present invention;

[0042] Figure 14 Block diagram of the device for optical biopsy of biological tissues provided by the embodiments of the present invention;

[0043] Figure 15 Structure diagram of the electronic device provided by the embodiments of the present invention. Detailed implementation manners

[0044] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0045] The method and device for optical biopsy of biological tissues according to the embodiments of the present invention will be described below with reference to the accompanying drawings. Regarding the CT scan mentioned in the above background technology as a common means for early screening of lung cancer, it can only present the approximate location and size information of tumors. For example, in actual clinical applications, for early lung nodules with small diameters and atypical shapes, it is difficult for CT scans to accurately judge their nature, which easily leads to missed diagnoses or misdiagnoses, causing patients to miss the best treatment opportunity. Although traditional endoscopes can directly observe the morphology of tumors in the trachea, their imaging depth and accuracy are seriously insufficient. When facing deep airway or micro-lesions, traditional endoscopes cannot clearly display the details of the lesions, making it difficult to provide substantial assistance for doctors' diagnostic processes. Often, patients need to undergo puncture biopsy to obtain pathological results. However, puncture biopsy is an invasive operation that may cause many complications, such as bleeding and pneumothorax. In severe cases, it may even lead to the death of patients. The present invention provides a method for optical biopsy of biological tissues. In this method, optical coherence tomography technology is used to perform imaging analysis on the target lumen, and a corresponding tissue biopsy image is generated by means of a cycle generative adversarial network model assisted by contrastive learning. Thus, the problems that endoscopy has certain limitations, it is difficult to accurately judge whether there is a lesion area through tissue images, and it needs to be combined with pathological examinations, which easily causes physical trauma to patients are solved.

[0046] Specifically, Figure 1 FIG. is a schematic flowchart of a method for optical biopsy of biological tissues provided by an embodiment of the present invention.

[0047] As Figure 1 shown, the method for optical biopsy of biological tissues includes the following steps:

[0048] In step S101, based on endoscopic optical coherence tomography technology, a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of the target lumen is obtained.

[0049] In some embodiments, obtaining a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of the target lumen based on endoscopic optical coherence tomography technology includes:

[0050] Based on endoscopic optical coherence tomography technology, an endoscope is set on a preset frequency-domain optical coherence tomography device;

[0051] The preset frequency-domain optical coherence tomography device emits visible light or near-infrared light, and the lumen tissue of the target lumen is scanned through the endoscope to obtain a two-dimensional cross-sectional image or a three-dimensional stereoscopic image.

[0052] Specifically, endoscopic optical coherence tomography technology is a non-invasive, high-resolution, non-labeled three-dimensional medical imaging technology that uses visible light or near-infrared light and the interference principle of light to perform three-dimensional tomographic imaging of tissues.

[0053] In the actual implementation process, the preset frequency-domain optical coherence tomography device includes a swept-source optical coherence tomography device (Swept-source OCT) and a spectral-domain optical coherence tomography device (Spectral-domain OCT). An endoscope is respectively arranged on the swept-source optical coherence tomography device and the spectral-domain optical coherence tomography device. Among them,

[0054] As Figure 2 shown, the light emitted by the swept light source of the swept-source optical coherence tomography device is divided into two beams by a coupler (such as a 90 / 10 splitting ratio). The stronger one of the two beams of light enters the sample arm, and the weaker one enters the reference arm. The light in the sample arm passes through a polarizer and reaches the motion module. The motion module is composed of a displacement motor and a scanning motor and can achieve three-dimensional scanning. The endoscope is connected to an optical fiber slip ring and is driven by a motor to rotate, so as to achieve distal scanning. The light in the reference arm passes through a delay line and reaches a mirror for reflection. The backscattered light from the sample arm and the reflected light from the reference arm interfere at the coupler (such as a 50 / 50 splitting ratio), are detected by a double-balanced photodetector, and finally form an image at the computer to obtain a two-dimensional cross-sectional image or a three-dimensional stereoscopic image;

[0055] As Figure 3 shown, the light beam emitted by the superluminescent light-emitting diode of the spectral-domain optical coherence tomography device is split by an optical fiber coupler. A part of the light beam enters the reference optical path, and a part of the light beam enters the sample optical path. The light beam entering the reference optical path is reflected at the mirror after passing through an optical delay line, passes through the optical delay line again, and returns to the optical fiber coupler. The light beam entering the sample optical path passes through a polarization controller and reaches the motion module. The motion module is composed of a displacement motor and a scanning motor and can achieve three-dimensional scanning to obtain a two-dimensional cross-sectional image or a three-dimensional stereoscopic image. At the same time, the light beam in the reference optical path and the scattered light beam in the sample optical path interfere in the coupler. Because the light source emits light beams of different wavelengths at the same time, there are multiple wavelengths of interference light. Therefore, a grating is used to spatially separate the interference light of different wavelengths and make it project onto different positions of a linear array or area array camera, so as to be able to detect the intensities of interference light of different wavelengths simultaneously.

[0056] Furthermore, a miniaturized endoscopic probe is provided on the endoscope. A lateral distal scanning probe based on an objective lens can be selected, or a lateral distal scanning probe composed of a series of fused optical fibers can be selected. The lateral distal scanning probe avoids the influence of a large size caused by integrating a motor at the probe end, and the lateral view imaging can scan the lumen structure 360° more, avoiding omission of areas.

[0057] Among them, as Figure 4 and 5As shown, the objective-based lateral distal scanning probe is connected by a single-mode fiber to a self-focusing lens to achieve the focusing of the laser. The converged laser is reflected by a 40° mirror to achieve lateral imaging. The entire probe structure is wrapped in a spring outer tube and a sleeve for protection. The lateral distal scanning probe composed of a series of fused fibers adopts a pure fiber structure. The single-mode fiber is fused with a coreless fiber for beam expansion and then fused with a self-focusing fiber for focusing. The converged laser is reflected by a 40° mirror to achieve lateral imaging. The entire probe structure is wrapped in a spring outer tube and a sleeve for protection.

[0058] Furthermore, in the design of the objective-based lateral distal scanning probe and the lateral distal scanning probe composed of a series of fused fibers, the focusing element is selected as a self-focusing lens or a self-focusing fiber. This structure can focus the light emitted by the fiber, thereby reducing the influence of reflection and obtaining better medical images. However, this structure needs to be aligned during assembly. In order to obtain better imaging effects in biological experiments, a mirror needs to be placed in front of the object side in advance to simulate the sample, so as to adjust the distance between the fiber and the lens.

[0059] As Figure 6 shown, a gradient refractive index lens is placed on the surface of the endoscopic probe. Among them, the gradient refractive index lens has unique optical properties, and its refractive index shows a gradient distribution. This enables the light to undergo continuous refraction according to the change of the refractive index when propagating in the lens, thereby achieving the focusing effect on the light beam. After the light emitted by the light source enters the lens, it will propagate along the path of the gradient refractive index and finally form a focused spot on the lumen tissue. This focusing method can provide relatively stable and efficient focusing performance, and can effectively improve the clarity and resolution of imaging of the lumen tissue.

[0060] As Figure 7 shown, a metasurface lens composed of a scatterer array is integrated in the endoscopic probe to utilize the unique ability of the metasurface lens to modulate the phase of the incident light at the sub-wavelength level, and can accurately image the point light source at a specific position to the diffraction-limited spot at the target position. Among them, the building unit of the metasurface lens is an amorphous silicon nanocolumn. The height of these nanocolumns is uniformly 750nm and they are arranged in a square lattice with a unit cell size of 400nm in an orderly manner. By changing the diameter of the nanocolumns, the phase given to the light can be flexibly controlled, thereby realizing the precise shaping of the light wavefront. This focusing method can still maintain high-resolution imaging outside the Rayleigh range of the incident light, successfully expanding the imaging depth of focus. This precise focusing performance greatly improves the clarity and accuracy of imaging.

[0061] In step S102, the two-dimensional cross-sectional image or three-dimensional stereoscopic image is converted by a pre-constructed cycle generative adversarial network model assisted by contrastive learning to generate the tissue pathological image of the target lumen.

[0062] In some embodiments, a two-dimensional cross-sectional image or a three-dimensional stereoscopic image is transformed by a pre-constructed cycle generative adversarial network model assisted by contrast learning to generate a histopathological image of a target lumen, including:

[0063] Training a cycle generative adversarial network model assisted by contrast learning using a preset unpaired dataset, and using cycle consistency loss to relieve the dependence on paired training data to obtain a cycle generative adversarial network model with enhanced contrast learning of image local features;

[0064] Using the cycle generative adversarial network model with enhanced contrast learning of image local features to transform a two-dimensional cross-sectional image or a three-dimensional stereoscopic image to generate a histopathological image of a target lumen.

[0065] It should be noted that in the field of medical images, OCT images (i.e., two-dimensional cross-sectional images or three-dimensional stereoscopic images, the following descriptions will all use this) and H&E stained images (i.e., histopathological images, the following descriptions will all use this) have different characteristics and application scenarios. Converting OCT images to H&E images is of great significance for medical diagnosis and research.

[0066] Based on the original CycleGAN, the cycle generative adversarial network assisted by contrast learning introduces a contrast learning mechanism. The traditional CycleGAN model can achieve image style transfer without relying on paired data. As Figure 8 shown, it can learn the mapping relationship between images on a dataset without precise pairing, greatly expanding its application in data-scarce or difficult-to-pair scenarios. However, there may be problems with insufficient retention of details and structures in the generated images. Contrast learning helps the model learn richer image content information by exploring the differences and similarities between local features of images. For example, when processing lumen tissue images, contrast learning can help the model better distinguish the subtle structural differences between normal and diseased tissue regions, making the generated H&E images more accurate in reflecting histopathological features.

[0067] When actually applied to medical images, for example, when processing OCT images of lung lumen tissues, the lung lumen tissues present specific hierarchical structures and optical reflection characteristics under OCT images, while H&E staining images highlight the microscopic morphology and staining characteristics of cells and tissues. The cycle generative adversarial network model assisted by contrastive learning in the embodiments of the present invention is trained with a large amount of OCT and H&E image data of lung lumen tissues of different individuals to learn the potential mapping relationship between the two. During the training process, the cycle consistency loss is used to ensure the reversibility of image conversion and avoid generating meaningless images; contrastive learning is used to analyze and learn the local features of images, so that when the generator generates H&E images, it can better retain the key structural information in the OCT images and at the same time generate cell and tissue features that conform to the style of H&E images, thereby achieving more effective image conversion and assisting medical professionals in more accurate pathological analysis.

[0068] As Figure 9 shown, the cycle generative adversarial network model assisted by contrastive learning consists of two sets of traditional GAN networks. The two systems respectively learn different transformation functions and simultaneously transform the input and output samples, enabling mutual transformation between the transformed and the source data to be transformed, so that unpaired datasets can be used for model learning.

[0069] In the conversion from OCT to H&E images, let X represent the OCT image data domain and Y represent the H&E image data domain. The generator of the first set of generative adversarial networks is G and the discriminator is D Y , where G is responsible for learning from OCT images and generating data similar to H&E images, and D Y is used to determine whether the data generated by G is consistent with the real H&E images in terms of features and distribution; in the second set of generative adversarial networks, the generator F learns from H&E images to generate data similar to OCT images, and the discriminator D X then determines whether the data generated by F belongs to the OCT image domain. As Figure 9 shown, overall, in the circular structure of the cycle generative adversarial network model assisted by contrastive learning, the two sets of generative adversarial networks transform data with each other and are respectively discriminated, so as to achieve the purpose of unsupervised learning.

[0070] As Figure 10As shown, the generator network structure of the cycle generative adversarial network model assisted by contrastive learning is also different from the convolutional neural network used by the discriminator. Residual blocks are added, that is, a transformation layer is added to the convolutional neural network. The use of the transformation layer makes the content output by the neural network ultimately directly added to the original input, so that the network can change as little as possible on the image. At the same time, a large amount of useless information is not connected to the neural network, that is, it skips the neural network, enabling the neural network to learn the features of the image more deeply.

[0071] As Figure 11 shown, in the loop structure of the cycle generative adversarial network model assisted by contrastive learning, the entire network consists of the mapping G: X → Y and the discriminator D Y and the mapping F: Y → X and the discriminator D X which are composed of 2 groups of generative adversarial networks.

[0072] Theoretically speaking, based on the above structure, after a large amount of learning and iteration, the CycleGAN model based on the cycle generative adversarial network can always learn the mapping relationship between X and Y. However, there are still potential mapping situations that do not meet expectations in the above structure. That is, although the generator is designed so that the data generated during continuous learning needs to be able to deceive the discriminator more and more, when the generator discovers that no matter what the input data x is, as long as the generated image is more like y, it can deceive the discriminator more easily, which may lead to the generator starting to generate only data that is the same as that in the Y domain.

[0073] Therefore, the embodiment of the present invention considers solving the minimax problem of the CycleGAN model based on the cycle generative adversarial network, and its optimization function is as follows:

[0074]

[0075] In the formula, there are 2 optimization processes for this minimax problem. On the one hand, fixing the generators G and F, the discriminators D X and D Y need to be trained to maximize the generated loss function, which means that the discriminator can distinguish as effectively as possible which samples are real and which are generated by the generator. On the other hand, fixing the discriminators D X and D Y , the generators G and F are trained to minimize the loss function, which means that the generators need to "deceive" the discriminator that has been trained as much as possible and generate fake data that is similar enough to the real data.

[0076] The CycleGAN model proposed in the embodiment of the present invention has four loss functions: adversarial loss (L GAN ), cycle consistency loss (L cyc) The segmentation loss (L cl ) and the feature map loss (L perceptual ). The data distributions of the OCT domain and the H&E domain are represented by and respectively.

[0077] The mapping function from OCT to H&E is denoted by G H&E , and the inverse mapping function is denoted by G OCT . Similarly, the discriminators are denoted by D H&E and D OCT . In the embodiments of the present invention, x is used to represent the input OCT or H&E image, and the adversarial loss is defined as follows:

[0078]

[0079]

[0080] where, E H&E (x) and E OCT (x) represent the expected values of data and data respectively. In the context of GAN, G H&E attempts to generate realistic H&E stained images, while D H&E attempts to distinguish the transformed image G H&E (x) from the real H&E stained image.

[0081] In the embodiments of the present invention, L GAN is used to train the generator and the discriminator. As shown in formula (2), L GAN consists of and . The discriminator aims to minimize by determining whether the input image is real using the first term in formula (3) and whether the transformed image is fake using the second term in formula (3). On the contrary, the generator aims to minimize through the discriminator based on formula (4). Since the label of the real image is 1 and the label of the transformed image is 0, the discriminator attempts to assign a value close to 1 to the real image and a value close to 0 to the transformed image. The adversarial loss L GAN (G H&E , D H&E ) of the mapping function from H&E to OCT can be defined accordingly.

[0082] Traditional image generation methods usually rely heavily on a large amount of accurately annotated paired image data. For example, in the field of medical imaging, if you want to convert ultrasound images into magnetic resonance imaging (MRI) images, you need to obtain paired samples of the same part and the same time, and annotate them carefully. This not only requires a lot of manpower, material resources and time, but also in actual operation, due to limitations of imaging conditions, individual differences among patients and other factors, it is difficult to collect sufficient and reliable paired data.

[0083] The contrastive learning-assisted cyclic generative adversarial network model adopted in the embodiment of the present invention breaks this limitation. It can directly process unlabeled data and mine the potential structure and patterns within the data. In the image generation scenario of the embodiment of the present invention, the contrastive learning-assisted cyclic generative adversarial network model can be trained using massive amounts of unpaired image data, greatly broadening the source and scale of the data. This approach does not require a cumbersome data labeling process, reduces the cost of data collection and preprocessing, and also avoids errors caused by inaccurate or inconsistent labeling, providing a more efficient and flexible solution for image generation tasks.

[0084] The embodiment of the present invention adopts a cyclic generative adversarial network model assisted by contrastive learning to achieve unsupervised image generation, the core of which lies in the adversarial training between the generator and the discriminator and the cycle consistency constraint.

[0085] In the adversarial training process, the generator and the discriminator form a system of mutual competition. H&E Try to learn the mapping relationship between OCT image domain and H&E image domain, and generate H&E images G that are as realistic as possible H&E (x). Discriminator D y The task is to distinguish the generated image G H&E (x) and the real H&E image. The generator and the discriminator form a dynamic balance by continuously iterating and optimizing their own parameters. The generator strives to generate images that can deceive the discriminator, while the discriminator continuously improves its discrimination ability. This adversarial training mechanism enables the generator to gradually capture the distribution characteristics of the target H&E image domain, thereby generating images with similar distribution.

[0086] Cycle consistency constraint is another key element of unsupervised learning based on contrastive learning-assisted cyclic generative adversarial network model, which requires that the generated image can be restored to the original image through the reverse generator. Specifically, for the input image x, after the generator G H&E Get G H&E (x), and then through the reverse generator G OCT Get G OCT [G H&E(x)]. The cyclic consistency loss L cyc (G H&E , G OCT ) measures the difference between G OCT [G H&E (x)] and the original OCT image. By minimizing the cyclic consistency loss, the generator can learn a more meaningful mapping relationship and avoid generating meaningless or distorted images. This constraint ensures that the generated H&E image is not only similar in appearance to the target H&E image but also semantically consistent with the OCT image, thus improving the quality and reliability of image generation.

[0087] The main purpose of the cyclic consistency loss is to ensure that x = G OCT [G H&E (x)], where the cycle consistency loss is defined as:

[0088] L cyc (G H&E , G OCT ) = E OCT {‖G OCT [G H&E (x)] - x‖1}

[0089] + E H&E {‖G H&E [G OCT (x)] - x‖1}(5)

[0090] where ‖‖1 represents the L1 norm.

[0091] In terms of feature analysis and extraction, as Figure 12 shown, the embodiments of the present invention select the VGG series models (with the number of network layers between 16 and 19) in the convolutional neural network model. The VGG series network structure is simple and is constructed by consecutive 3×3 small convolutional kernels and pooling layers. This structure combines features by scanning the picture multiple times with fewer convolutional kernels, which can improve the performance of the neural network while reducing the number of parameters compared to using large convolutional kernels.

[0092] In the VGG16 network, the convolutional layer is responsible for feature analysis and extraction of the input data. Consecutive 3×3 small convolutional kernels are connected in series to form a convolutional kernel with a larger receptive field, enhancing the feature extraction ability without increasing the computational amount. The pooling layer compresses the data features and reduces the overfitting rate of the model. The Relu activation function is added after each convolutional layer to introduce non-linear features for model training and alleviate the problem of gradient disappearance.

[0093] The Flatten layer flattens the image features into one-dimensional data and inputs it into the fully connected layer. The neurons in the fully connected layer learn the optimal weight and bias values through forward and backward propagation during the training process, and perform a higher-level feature representation on the input data. The Dropout layer reduces the interaction between specific or strongly correlated nodes by making the activation value of a certain neuron stop working with a certain probability P, thereby improving the generalization performance of the model.

[0094] In the context of the continuous development of current medical image conversion technology, traditional methods face many challenges when dealing with complex medical image data, such as difficulties in accurately capturing subtle features in images and insufficient adaptability to differences between different modality images. To break through these technical bottlenecks and improve the accuracy and efficiency of medical image conversion, the present invention implementation introduces contrastive learning. By maximizing or minimizing the similarity metric between data features, the model is guided to learn a more discriminative feature table, such as Figure 13 As shown, contrastive learning can effectively enhance the model's ability to retain the content and structure of medical images, while improving the model's understanding and conversion ability of image content, thereby improving the accuracy and reliability of image conversion.

[0095] In actual operation, contrastive learning is carried out based on a carefully designed sample selection strategy. Representative samples are selected from the original OCT images of the lumen tissue and the corresponding generated H&E images. Some image regions are selected as samples in the original OCT images, and at the same time, regions at the corresponding positions are selected as contrast samples in the generated H&E images. This sample selection method closely focuses on the pathological features of biological tissues and can effectively guide the model to learn the differences in image features between normal tissues and diseased tissues.

[0096] The key to contrastive learning lies in feature extraction and contrastive loss calculation. A specific feature extraction function implemented based on the multi-layer structure of the convolutional neural network is used to process the selected samples. Through convolutional operations, the sample image patches are transformed into feature vectors containing rich texture, structure, and other information. These feature vectors are substituted into the contrastive learning loss function to calculate the contrastive loss, and its definition is as follows:

[0097]

[0098] Among them, N and M respectively represent the number of feature regions selected from the source image and the generated image, x represents the source image, represents the generated target image, f i and f j are functions used to extract image features, D is a distance metric function used to measure the difference between two feature vectors, such as cosine distance or Euclidean distance, w ijis a weight coefficient assigned according to feature importance or other strategies, used to adjust the contribution degree of different feature regions to loss calculation.

[0099] In the task of lumen tissue image conversion, through this contrast loss function, the model can carefully compare the features of the original OCT image and the generated H&E image. For example, when x is the original OCT image of the lumen tissue, and y is the converted H&E image, f i (x) corresponds to extracting the feature vector of a certain region in the original OCT image, and f(y) corresponds to the feature vector of the same position or related region in the generated image. Using the contrast loss function to calculate the difference between the two can quantify this degree of difference, so as to more accurately present the lumen tissue features in the generated H&E image and optimize the image conversion effect.

[0100] At the same time, the embodiment of the present invention also introduces a perceptual loss function (L perceptual ), which uses a pre-trained neural network to extract the perceptual features of the image. By calculating the difference between the generated image and the real image in these perceptual features, the quality of the generated image is optimized from the perceptual level. Its expression is:

[0101]

[0102] where L represents the number of layers used to calculate the perceptual loss in the pre-trained network, n l is the number of elements in the feature map of the l-th layer, φ l represents the feature extraction function of the l-th layer of the pre-trained network, x is the real image, and y is the generated image. By minimizing the perceptual loss, the generated histopathological image is closer to the real histopathological image in terms of visual effect, can provide more intuitive and accurate pathological information for doctors, and assist doctors in making more reliable diagnoses.

[0103] In summary, in order to ensure real image translation, the overall loss function expression of the cycle generative adversarial network model assisted by contrast learning for training is:

[0104] L total (G H&E ,G OCT ,D H&E ,D OCT )=L GAN (G H&E ,D H&E )+L GAN (G OCT ,D OCT )

[0105] +λ·L cyc(G H&E ,G OCT ) + α·L cl (G H&E ,G OCT ) + β·L perceptual (G H&E ,G OCT )

[0106] Wherein, L total is the overall loss function, G H&E is the histopathological image generator, G OCT is the optical coherence tomography generator, D H&E is the histopathological image discriminator, D OCT is the optical coherence tomography discriminator, L GAN is the adversarial loss function, L cyc is the cycle consistency loss function, L cl is the image feature contrast loss function, L perceptual is the perceptual loss function, and λ, α, and β represent coefficients that control the correlation of different losses.

[0107] That is to say, based on the above overall loss function, the embodiments of the present invention use a preset unpaired dataset to train a cycle generative adversarial network model assisted by contrast learning, so as to obtain an image local feature contrast learning enhanced cycle generative adversarial network model, and then convert a two-dimensional cross-sectional image or a three-dimensional stereoscopic image according to the image local feature contrast learning enhanced cycle generative adversarial network model to generate a histopathological image with rich image features of the target lumen.

[0108] In summary, the method for optical biopsy of biological tissues proposed according to the embodiments of the present invention has the following beneficial effects:

[0109] (1) By using optical coherence tomography technology to image and analyze the lumen tissue, and generating corresponding tissue biopsy images with the help of a cycle generative adversarial network model assisted by contrast learning, there is no need to puncture the patient, avoiding physical trauma and potential risks caused by puncture, and providing a safer screening option for the patient;

[0110] (2) Through the long-distance endoscopic optical coherence tomography system, it is possible to more clearly capture details such as the morphology and structure of small lung lesions, assist doctors in diagnosing the lung condition, enabling patients to receive treatment at an early stage of the disease, and greatly improving the treatment success rate and the quality of life of the patients;

[0111] (3) In terms of clinical application, once this technology is widely promoted, it will have an important impact on the early screening of lung cancer. Early detection of lung cancer is crucial for improving the survival rate of patients and improving the prognosis;

[0112] (4) From the perspective of the development of the medical industry, the application of this technology will promote the innovation of lung cancer screening technology and encourage the development of medical equipment in a more accurate and non-invasive direction;

[0113] (5) From a social perspective, it can effectively reduce the heavy economic burden of lung cancer treatment on families and society, improve the overall health level and quality of life of society, and have significant social benefits.

[0114] Next, the device for optical biopsy of biological tissue according to the embodiment of the present invention will be described with reference to the accompanying drawings.

[0115] Figure 14 A schematic block diagram of a biological tissue optical biopsy device provided by an embodiment of the present invention.

[0116] like Figure 14 As shown, the biological tissue optical biopsy device 140 includes: an image acquisition module 1401 and an image conversion module 1402 .

[0117] The image acquisition module 1401 is used to acquire a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of the target lumen based on endoscopic optical coherence tomography technology. The image conversion module 1402 is used to convert the two-dimensional cross-sectional image or the three-dimensional stereoscopic image through a pre-built contrast learning-assisted recurrent generative adversarial network model to generate a histopathological image of the target lumen.

[0118] In some embodiments, the acquisition image module 1401 includes:

[0119] A setting unit, used for setting an endoscope on a preset frequency domain optical coherence tomography device based on endoscopic optical coherence tomography technology;

[0120] The scanning unit is used to use a preset frequency domain optical coherence tomography device to emit visible light or near-infrared light, and scan the lumen tissue of the target lumen through an endoscope to obtain a two-dimensional cross-sectional image or a three-dimensional stereoscopic image.

[0121] In some embodiments, the endoscope is provided with a gradient refractive index lens or a metasurface lens composed of an integrated scatterer array to focus visible light or near-infrared light onto the luminal tissue of the target lumen.

[0122] In some embodiments, the image conversion module 1402 includes:

[0123] A training unit, used for training a contrastive learning-assisted cyclic generative adversarial network model using a preset unpaired data set, and removing the dependence on paired training data using a cycle consistency loss, so as to obtain an image local feature contrastive learning enhanced cyclic generative adversarial network model;

[0124] A conversion unit is configured to enhance a cycle generative adversarial network model to convert a two-dimensional cross-sectional image or a three-dimensional stereoscopic image by using image local feature contrast learning, so as to generate a histopathological image of a target lumen.

[0125] In some embodiments, the preset unpaired dataset includes two-dimensional cross-sectional images and three-dimensional stereoscopic images of target lumens of different individuals, or three-dimensional stereoscopic images and histopathological images.

[0126] In some embodiments, the expression of the overall loss function of the cycle generative adversarial network model assisted by contrast learning is:

[0127] L total (G H&E ,G OCT ,D H&E ,D OCT ) = L GAN (G H&E ,D H&E ) + L GAN (G OCT ,D OCT )

[0128] + λ·L cyc (G H&E ,G OCT ) + α·L cl (G H&E ,G OCT ) + β·L perceptual (G H&E ,G OCT )

[0129] Wherein, L total is the overall loss function, G H&E is the histopathological image generator, G OCT is the optical coherence tomography generator, D H&E is the histopathological image discriminator, D OCT is the optical coherence tomography discriminator, L GAN is the adversarial loss function, L cyc is the cycle consistency loss function, L cl is the image feature contrast loss function, L perceptual is the perceptual loss function, and λ, α, and β represent coefficients for controlling the correlation of different losses.

[0130] It should be noted that the foregoing explanation of the method embodiments of optical biopsy of biological tissues also applies to the device for optical biopsy of biological tissues in this embodiment, and details are not described herein again.

[0131] The device for optical biopsy of biological tissue proposed according to the embodiments of the present invention has the following beneficial effects:

[0132] (1) By using optical coherence tomography technology to image and analyze the tissue in the lumen, and generating corresponding tissue biopsy images with the help of a cycle generative adversarial network model assisted by contrastive learning, there is no need to puncture the patient, avoiding physical trauma and potential risks caused by puncture, and providing a safer screening option for patients;

[0133] (2) Through the long-distance endoscopic optical coherence tomography system, it is possible to more clearly capture detailed information such as the morphology and structure of small lung lesions, assist doctors in diagnosing the lung condition, enabling patients to receive treatment at an early stage of the disease, and greatly improving the treatment success rate and the quality of life of patients;

[0134] (3) In terms of clinical application, once this technology is widely promoted, it will have an important impact on the early screening of lung cancer. Early detection of lung cancer is crucial for improving the survival rate of patients and improving the prognosis;

[0135] (4) From the perspective of the development of the medical industry, the application of this technology will promote the innovation of lung cancer screening technology, and prompt the research and development of medical equipment to develop in a more precise and non-invasive direction;

[0136] (5) Socially, it can effectively reduce the heavy economic burden brought by lung cancer treatment to families and society, improve the overall health level and quality of life of society, and has significant social benefits.

[0137] Figure 15 The following is a schematic structural diagram of the electronic device provided by the embodiments of the present invention. The electronic device may include:

[0138] A memory 1501, a processor 1502, and a computer program stored on the memory 1501 and executable on the processor 1502.

[0139] When the processor 1502 executes the program, it implements the method for optical biopsy of biological tissue provided in the above embodiments.

[0140] Furthermore, the electronic device further includes:

[0141] A communication interface 1503 for communication between the memory 1501 and the processor 1502.

[0142] The memory 1501 is used to store a computer program executable on the processor 1502.

[0143] The memory 1501 may include a high-speed RAM memory, and may also include non-volatile memory, such as at least one disk memory.

[0144] If the memory 1501, the processor 1502, and the communication interface 1503 are implemented independently, the communication interface 1503, the memory 1501, and the processor 1502 can be interconnected through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 15 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0145] Optionally, in a specific implementation, if the memory 1501, the processor 1502, and the communication interface 1503 are integrated on a single chip, the memory 1501, the processor 1502, and the communication interface 1503 can communicate with each other through an internal interface.

[0146] The processor 1502 may be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0147] The embodiments of the present invention also provide a computer program product, and when the computer program / instructions are executed by a processor, the method for optical biopsy of biological tissue as described above is implemented.

[0148] The embodiments of the present invention also provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the method for optical biopsy of biological tissue as described above is implemented.

[0149] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples", etc., mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0150] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise specifically defined.

[0151] Any process or method description shown in the flowchart or described in other ways herein can be understood to represent a module, segment, or portion of code including one or N executable instructions for implementing a customized logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0152] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definitional sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or used in conjunction with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part (electronic device) having one or N wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing as appropriate, and then storing it in a computer memory.

[0153] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0154] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the methods of the above embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0155] In addition, each functional unit in various embodiments of the present invention may be integrated into one processing module, may exist separately as individual physical units, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0156] The above-mentioned storage medium may be a read-only memory, a magnetic disk, an optical disc, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for optical biopsy of biological tissue, characterized in that: The following steps are involved: Based on endoscopic optical coherence tomography technology, a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of the target lumen is obtained; The two-dimensional cross-sectional image or the three-dimensional stereoscopic image is converted by a pre-constructed contrast learning-assisted recurrent generative adversarial network model to generate a tissue pathology image of the target lumen.

2. The method for optical biopsy of biological tissue according to claim 1, characterized in that: The method of obtaining a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of a target lumen based on endoscopic optical coherence tomography technology includes: Based on endoscopic optical coherence tomography technology, an endoscope is arranged on a preset frequency domain optical coherence tomography device; Visible light or near-infrared light is emitted by the preset frequency domain optical coherence tomography device, and the lumen tissue of the target lumen is scanned through the endoscope to obtain the two-dimensional cross-sectional image or the three-dimensional stereoscopic image.

3. The method for optical biopsy of biological tissue according to claim 2, characterized in that: The endoscope is provided with a gradient refractive index lens or a super surface lens composed of an integrated scatterer array to focus visible light or near infrared light onto the lumen tissue of the target lumen.

4. The method for optical biopsy of biological tissue according to claim 1, characterized in that: The converting of the two-dimensional cross-sectional image or the three-dimensional stereoscopic image by a pre-constructed contrast learning-assisted recurrent generative adversarial network model to generate a tissue pathological image of the target lumen comprises: The contrastive learning-assisted recurrent generative adversarial network model is trained using a preset unpaired data set, and the cycle consistency loss is used to remove the dependence on paired training data, so as to obtain an image local feature contrastive learning enhanced recurrent generative adversarial network model; The two-dimensional cross-sectional image or the three-dimensional stereoscopic image is converted by using the image local feature contrast learning enhanced recurrent generative adversarial network model to generate a tissue pathology image of the target lumen.

5. The method for optical biopsy of biological tissue according to claim 4, characterized in that: The preset non-paired data set includes a two-dimensional cross-sectional image and a three-dimensional stereoscopic image, or a three-dimensional stereoscopic image and a tissue pathology image, of different individual target luminal tracts.

6. The method for optical biopsy of biological tissue according to claim 1, characterized in that: The expression of the overall loss function of the contrastive learning-assisted cyclic generative adversarial network model is: L total (G H&E ,G OCT ,D H&E ,D OCT )=L GAN (G H&E ,D H&E )+L GAN (G OCT ,D OCT ) +λ·L cyc (G H&E ,G OCT )+α·L cl (G H&E ,G OCT )+β·L perceptual (G H&E ,G OCT ) Among them, L total is the overall loss function, G H&E is the tissue pathology image generator, G OCT For optical coherence tomography generator, D H&E is the tissue pathology image discriminator, D OCT is the optical coherence tomography discriminator, L GAN is the adversarial loss function, L cyc is the cycle consistency loss function, L cl is the image feature contrast loss function, L perceptual is the perceptual loss function, and λ, α, and β represent the coefficients that control the correlation of different losses.

7. A device for optical biopsy of biological tissue, characterized in that: include: An image acquisition module, used for acquiring a two-dimensional cross-sectional image or a three-dimensional stereoscopic image of a target lumen based on endoscopic optical coherence tomography technology; An image conversion module is used to convert the two-dimensional cross-sectional image or the three-dimensional stereoscopic image through a pre-built contrast learning-assisted recurrent generative adversarial network model to generate a tissue pathology image of the target lumen.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for optical biopsy of biological tissue according to any one of claims 1 to 6.

9. A computer program product, characterized in that When the computer program / instructions are executed by a processor, the method for optical biopsy of biological tissue described in any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method for optical biopsy of biological tissue as described in any one of claims 1 to 6.

Images